A Deep-Learning-Enabled Monitoring System for Ocular Redness Assessment

Abstract

<p>Ocular redness is a common symptom of numerous eye conditions and serves as an essential diagnostic indicator requiring accurate and timely assessment. However, the conventional manual evaluation of ocular redness is inherently subjective, inefficient, and error-prone due to inter-observer variability. To address these limitations, we present an automatic ocular redness monitoring system (AORMS) that utilizes deep learning models for objective and consistent quantification of ocular redness. In this work, we propose an approach for effectively classifying and monitoring ocular redness caused by subconjunctival hemorrhage (SCH) and conjunctivitis. To ensure robustness, we employ transfer learning and image processing techniques to maximize the utilization of a limited dataset comprising external eye photos. Additionally, a complete pipeline is implemented to facilitate the seamless integration of our system into clinical workflows. The proposed method achieved 98.3 % accuracy in class classification and 96.2 % accuracy in SCH area identification.<br></p>